Adrian ReidFounder of Enlightened Stock Trading

Should I Backtest My Trading Strategies Back To 1950

26 May 2025 · 2 min

Watch on YouTube

1:32

Adrian Reid advises choosing backtest horizons to match strategy timeframes: use as much data as possible but be mindful markets change over time. For long-term trend-following he prefers data back to ~1990 (or back to 1950 for US-only trend following) to capture varied market regimes; for short-duration mean-reversion systems he recommends using more recent data (about the last 10 years). He also emphasizes needing large sample sizes—hundreds or preferably thousands of trades—to get reliable backtest results.

Questions this video answers

Back to around 1990 (or further if available).

Back to 1950.

About the last 10 years (use more recent history).

Hundreds, if not thousands of trades (avoid samples with only 10, 50, or 100 trades).

Chapters

The further back you go, the more markets have changed. For some strategies, you want to use more recent history. For other strategies you want to go back further. I like to use as much data as possible, acknowledging that the further back you go, the more markets have changed. So for example, if I've got a long term trend following system, I'd love to see that works well all the way back to 1990. Or if it's a US trend following system, I can apply my US trend following system to data back to 1950. And the cool thing about that is, yeah, the markets have changed a lot. However, if you go all the way back then you've got super high inflation, falling inflation, high interest rates, low interest rates, all sorts of crazy things happening. You've got the October 1987 crash, so a really wide variety of market conditions which allows you to test whether or not the system will survive.
All of those different conditions. And the cool thing is our trend following systems do. Now if you applied some of our shorter term systems way back then, they don't work so well because the market dynamics have changed so much. So for a mean reversion system that trades only one or two days at a time, you might use the last 10 years of data. We just need to make sure we get enough data to have enough trades in our sample to get valid results, essentially. And we want hundreds, if not thousands of trades in our back test. We don't want 10, 50, 100 trades because that's not enough to really get a good sense for how profitable and stable the system is.